Research map / Label-efficient and robust learning / Recognition and long-tail
Noise is a Good Teacher: A Noise-Driven Framework for Robust Collaborative Perception
Chengyan Huang, Zhongxiang Zhao, Ziyao Zhang, Zihao Yang, Lin Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add multi-level pose noise injection (Module A) (ablation) | NoiseGT · V2X-Sim (pose noise 1.2/1.2) | AP@0.7: 0.577 -> 0.638 | Improved |
| add multi-level pose noise injection (Module A) (ablation) | NoiseGT · V2X-Sim (clean pose) | AP@0.7: 0.838 -> 0.824 | Hurt |
| add feature denoiser (Module B) (ablation) | NoiseGT · V2X-Sim (clean pose) | AP@0.7: 0.838 -> 0.851 | Improved |
| add feature denoiser (Module B) alone (ablation) | NoiseGT · V2X-Sim (pose noise 1.2/1.2) | AP@0.7: 0.577 -> 0.577 | No clear effect |
| combine modules A and B (ablation) | NoiseGT · V2X-Sim (pose noise 1.2/1.2) | AP@0.7: 0.638 -> 0.661 | Improved |
| replace fixed Gaussian smoothing with learned noise-driven denoiser (ablation) | NoiseGT FDM · V2X-Sim (pose noise 1.2/1.2) | AP@0.7: 0.613 -> 0.661 | Improved |
| add noise injection and denoise loss over architecture-only refinement (ablation) | NoiseGT FDM · V2X-Sim (pose noise 1.2/1.2) | AP@0.7: 0.654 -> 0.661 | Improved |
| add explicit denoising over noise augmentation only (ablation) | NoiseGT · V2X-Sim (pose noise 1.2/1.2) | AP@0.7: 0.612 -> 0.661 | Improved |
| add NoiseGT multi-level noise injection to RoCo solver | RoCo · V2X-Sim (pose noise 1.2/1.2) | AP@0.5: 0.703 -> 0.733 | Improved |
| swap PointPillars for SECOND backbone with NoiseGT vs CoAlign | NoiseGT · V2X-Sim SECOND (pose noise 1.2/1.2) | AP@0.5: 0.667 -> 0.720 | Improved |
| swap PointPillars for VoxelNet backbone with NoiseGT vs CoAlign | NoiseGT · V2X-Sim VoxelNet (pose noise 1.2/1.2) | AP@0.5: 0.683 -> 0.741 | Improved |
| use NoiseGT instead of CoDiff at matched one diffusion step | NoiseGT · DAIR-V2X (pose noise 1.2/1.2) | AP@0.5: 0.652 -> 0.691 | Improved |
Closest papers by meaning
- CoRA: A Collaborative Robust Architecture with Hybrid Fusion for Efficient Perception (AAAI 2026 · similarity 0.88)
- RCP-Bench: Benchmarking Robustness for Collaborative Perception Under Diverse Corruptions (CVPR 2025 · similarity 0.87)
- Enhance Vision-Language Alignment with Noise (AAAI 2025 · similarity 0.86)
- Multi-agent Collaborative Perception via Motion-aware Robust Communication Network (CVPR 2024 · similarity 0.85)
- What Makes Good Collaborative Views? Contrastive Mutual Information Maximization for Multi-Agent Perception (AAAI 2024 · similarity 0.84)
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
